Welcome back to the journal review. This is the October twenty twenty-six issue of the Journal of the American Academy of Dermatology, and we've got four pieces to get through today — two original research brief reports, and a letter-and-response pair that together make for one of the better methodology lessons I've seen in this journal in a while. Let's get into it.

First up is a brief report on global trends in melanoma incidence and disability burden among young adults, stratified by socioeconomic development level, using Global Burden of Disease data. This is a database-driven descriptive epidemiology study, essentially an ecological analysis rather than a patient-level cohort, and the gap it's addressing is straightforward: we know melanoma skews older, but it remains one of the more commonly diagnosed cancers in people age fifteen to thirty-nine, and we don't have a clean global picture of how that burden is shifting across different levels of economic development.

Methodologically, the authors pulled age-standardized incidence rates and disability-adjusted life year rates for this age band from nineteen ninety through twenty twenty-one out of the Global Burden of Disease Results Tool, then stratified everything by the Socio-demographic Index, which is a composite measure of income, education, and fertility used as a proxy for overall development — think the United States as high, Russia as high-middle, India as middle, Indonesia as low-middle, Kenya as low. The trend metric they leaned on is called the estimated annual percentage change, and it's worth pausing on what that actually is. It's essentially a log-linear regression — you take the yearly rates, log-transform them, and fit a straight line through time. The slope of that line, converted back out of log space, gives you an average annual percent change that smooths over year-to-year noise and lets you compare trend steepness across groups with very different baseline rates. It's the right tool here because the authors don't care about any single year's bump, they care about the overall trajectory across three decades, and a log-linear fit handles exponential-type growth or decay naturally. The obvious limitation, which they don't flag but is worth noting as my own read, is that this is purely descriptive — there's no adjustment for confounders, no patient-level data, just population-level rates moving over time.

So what did they find. Globally, incidence was essentially flat over thirty years — around one per hundred thousis young adults in both nineteen ninety and twenty twenty-one, and that overall trend wasn't statistically significant. High SDI regions had by far the highest absolute incidence, roughly five per hundred thousand, but also the flattest trend, essentially no significant change. The real story is in the lower-resource tiers: high-middle, middle, and low-middle SDI regions all showed statistically significant increases in incidence, with annual percentage increases in the range of two to two-and-a-half percent — genuinely steep trajectories, even though their absolute rates remain much lower than high SDI regions. Low SDI regions also showed a significant, if more modest, upward incidence trend. Disability burden told an interesting counter-story: it fell significantly worldwide and in high and high-middle SDI regions, but it actually rose significantly, albeit by less than one percent a year, in middle, low-middle, and even low SDI regions.

The authors' interpretation is that high SDI regions' high-but-flattening incidence likely reflects saturation — more awareness, more screening, more sun protection and treatment already baked in — while rising incidence paired with rising disability in lower-resource regions is a worse pattern, suggestive of a mix of improving case ascertainment but also under-diagnosis and later-stage presentation when people are caught at all. They're explicit about limitations — data gaps and registry incompleteness in low SDI regions, SDI being an imperfect proxy that doesn't capture every disparity in treatment or diagnostic access, no race or ethnicity data at all, and the fact that estimates run through twenty twenty-one so they capture pandemic-era disruption but nothing after it.

Practically, I'd call this interesting but not actionable at the level of an individual Mohs or dermatologic oncology practice — there's no new diagnostic or treatment signal here. Where it is useful is as ammunition for the bigger-picture conversation about global equity in melanoma care: if you're involved in outreach, policy, or international education efforts, this is a data point supporting investment in earlier detection infrastructure in rising-incidence, lower-resource settings rather than continued concentration of screening resources where incidence is already high but plateaued.

Second article: a pilot study titled increased dendritic cell density and altered melanin patterns distinguish melanoma from nevi on cellular-resolution full-field optical coherence tomography, or CRFF-OCT, framed explicitly as a step toward noninvasive diagnosis. This is a single-center, prospective, cross-sectional diagnostic pilot study, and the gap is the familiar one — dermoscopy and reflectance confocal microscopy are noninvasive but shallow, capped at around two hundred micrometers and largely horizontal in orientation, whereas this CRFF-OCT platform penetrates to about five hundred micrometers, images vertically the way histopathology does, and can reconstruct three-dimensionally.

The design makes sense for a first-in-concept study: they imaged biopsy-proven melanocytic lesions plus immediately adjacent perilesional skin as an internal, patient-specific control, then ran the images through an AI-based computer-aided detection platform that quantified epidermal dendritic cell metrics — count, size, density, distribution — and built a volumetric melanin map from the combined vertical and en face images, specifically to avoid the bias of judging pigment from a single two-dimensional slice. Only seventeen patients went into this, all Fitzpatrick skin type two, split into eleven melanomas — seven invasive, four in situ — and six nevi, a mixed bag of dysplastic, dermal, and Spitz lesions. For statistics, they used the Kolmogorov-Smirnov test, a t-test, and the Mann-Whitney U test. The sequencing there is a nice little teaching point: Kolmogorov-Smirnov is checking whether your data actually follow a normal distribution, and that answer determines which comparison test is valid downstream — if the data are normal, a t-test comparing group means is appropriate, but with a sample this small, data are often not well-behaved enough to trust a mean-based test, so Mann-Whitney U, which instead ranks all the observations and compares the rank sums between groups, becomes the safer choice because it doesn't assume any particular distribution shape and is less sensitive to outliers in a dataset of this size.

The findings: melanomas showed significantly higher epidermal dendritic cell density than both nevi and perilesional control skin. Melanin density was significantly higher in both melanomas and nevi compared to control skin, but — and this is the key nuance — there was overlap between the melanoma and nevus groups on melanin density alone, meaning pigment quantity by itself isn't a reliable discriminator. Where it got more interesting was granule-level architecture: mean melanin granule size was significantly larger in melanoma than in control skin, and inter-granule distance was significantly greatest in nevi, intermediate in melanoma, and smallest in controls — which tracks nicely with the known histologic picture of orderly benign nesting in nevi versus crowded, architecturally disorganized melanocytes in melanoma.

The authors are appropriately cautious. They call these findings preliminary, and the limitations are substantial for generalizability: seventeen patients, one center, every patient Fitzpatrick type two only, which matters enormously for a melanin-quantification tool since pigment metrics will behave very differently in darker skin types, heterogeneous lesion subtypes lumped together, and no formal slide-by-slide histopathologic correlation to anchor the imaging findings. They also note a practical bottleneck — image acquisition takes under five minutes, but the current AI analysis pipeline can take hours to days, which rules out any real-time bedside use right now.

My take for practice: this is a genuinely interesting proof-of-concept that pigment architecture and dendritic cell density can be quantified noninvasively and volumetrically, and it reinforces that multiparametric approaches beat any single pigment metric, which fits what we already know from RCM and dermoscopy. But it is not remotely practice-changing yet — too small, too narrow a skin type range, no histologic ground-truthing, and no real-time workflow. File it as a technology to watch for a larger multicenter validation study, not something to discuss with patients yet.

Now the last piece, which is really two pieces in conversation with each other — a letter titled time as a substantial confounder in evaluating integrated multidisciplinary care for advanced cutaneous squamous cell carcinoma, and the response from the original study's authors. Neither of these is original research with its own data, so there's no methods-results-limitations scaffold to walk through — this is correspondence, and the value here is entirely in the methodological argument, which is excellent and worth sitting with.

The backstory: a prior JAAD study by Ran and colleagues found that after their institution rolled out an integrated multidisciplinary care model for high-risk cutaneous squamous cell carcinoma in twenty twenty, time to treatment initiation dropped by seventeen days compared to the pre-2020 historical cohort, along with fewer pre-treatment clinic visits. Wang, Li, and Pang write in to raise a sharp point: in that design, exposure to integrated care is perfectly collinear with calendar time. Everyone before April twenty twenty is "non-integrated," everyone after is "integrated." That means time itself becomes an unmeasurable stand-in for every other thing that was also changing across that same window — PD-1 inhibitor approvals, growing use of neoadjuvant immunotherapy, the COVID-19 pandemic's disruption of care pathways, evolving clinician experience, shifting patient behavior. Their core statistical point is one worth internalizing generally: you cannot statistically adjust for time as an ordinary covariate when your exposure variable is defined by a single point in time, because any unmeasured confounder that trends with calendar time is, by construction, inseparable from your intervention. They note that historical-control, before-and-after designs are only trustworthy when the treatment effect is dramatic enough to swamp plausible background drift — and a seventeen-day shift, while real, isn't in that category. Their proposed fixes are the standard quasi-experimental upgrades: an interrupted time series design, which plots the outcome across many time points before and after the intervention and asks whether there's a distinct jump or slope change right at the intervention date, over and above whatever secular trend was already happening — versus a difference-in-differences design, which would compare the change over time at this institution to the change over time at a similar institution that never adopted integrated care, using that second site's trajectory as the counterfactual for what would have happened anyway, under the assumption that both sites would have trended similarly without the intervention.

Ran and colleagues' response is measured and genuinely useful. They agree that exposure is aligned with time, and they explain why a contemporaneous within-institution comparison group wasn't feasible — systems-level clinic changes get rolled out to everyone at once, and cross-institution comparisons bring their own confounding from differing infrastructure and workflow. But they push back substantively in two ways. First, they ran a subgroup analysis restricted to patients who did not receive neoadjuvant immunotherapy, specifically to address the concern that the whole effect was just an artifact of a new drug class becoming available — and even in that subgroup, time to treatment was still significantly shorter in the integrated-care cohort, twenty-one days versus thirty-six days. Second, on the pandemic point, they note that COVID-19 is well documented to have delayed skin cancer care generally, which means its effect should have biased the post-2020 integrated group toward a longer time to treatment, not a shorter one — so the fact that they still found a shorter time to treatment despite a pandemic-related headwind is, if anything, a point in favor of a real effect rather than against it. They do concede the central point: this design cannot demonstrate that integrated care alone, as opposed to the bundle of changes happening alongside it, caused the improvement, and they explicitly call for future interrupted time series or difference-in-differences studies to settle it.

There's no results-and-limitations section to recite for correspondence like this, so the real takeaway is about how to read care-delivery literature. For a practicing Mohs surgeon or dermatologic oncologist, this exchange isn't going to change what you do tomorrow, but it's a sharp reminder to be skeptical of before-and-after institutional studies whenever the intervention date lines up with a known era of broader change — new drug approvals, a pandemic, workflow overhauls — because conventional statistical adjustment genuinely cannot rescue that kind of collinearity. The substantive clinical idea, that integrating diagnostics, scheduling, and multidisciplinary case conferences up front shortens time to treatment for advanced cutaneous squamous cell carcinoma, remains plausible and is operationally reasonable to pursue regardless of this debate. But if you're citing this seventeen-day figure to justify building your own multidisciplinary pathway, cite it as suggestive evidence of benefit from care coordination, not as proof that the specific model caused the specific number.

That wraps our four pieces for this episode — a sobering global look at melanoma's shifting burden in younger patients, an early but promising optical imaging approach to noninvasive melanocytic lesion assessment, and a methodology masterclass hiding inside a two-letter exchange about multidisciplinary care. Thanks for listening, and we'll see you next month.